autoresearch-vibepi-000253-cucumber-grab
收藏资源简介:
该数据集是一个机器人操作数据集,使用LeRobot工具创建,专注于收集机器人状态和视觉观测。它包含多模态数据:机器人状态观测(包括肩部平移、肩部提升、肘部弯曲、腕部弯曲、腕部旋转、夹爪位置和倾斜位置等7个浮点数值)、来自三个摄像头的视觉观测(基础摄像头、顶部摄像头和腕部摄像头,均以20fps录制480x640分辨率的RGB视频)、对应的机器人动作(7维浮点数组),以及时间戳、帧索引、片段索引、任务索引等元数据。数据集总共有1个任务片段,包含169帧数据,数据以Parquet文件格式存储,视频以MP4格式存储,机器人类型为vibeboard_follower_tilt。该数据集适用于机器人学习、强化学习和视觉-动作映射等研究任务。
This dataset is a robot operation dataset created using the LeRobot tool, focusing on the collection of robot state and visual observations. It contains multimodal data: robot state observations (including 7 floating-point values such as shoulder translation, shoulder elevation, elbow flexion, wrist flexion, wrist rotation, gripper position, and tilt position), visual observations from three cameras (base camera, top camera, and wrist camera, all recording 480x640 resolution RGB video at 20fps), corresponding robot actions (7-dimensional floating-point arrays), as well as metadata such as timestamps, frame indices, segment indices, and task indices. The dataset consists of 1 task segment with 169 frames of data, stored in Parquet file format for data and MP4 format for videos, with the robot type being vibeboard_follower_tilt. It is suitable for research tasks in robot learning, reinforcement learning, and vision-action mapping.
数据集概述
- 数据集名称: autoresearch-vibepi-000253-cucumber-grab
- 许可证: Apache-2.0
- 任务类别: 机器人学(Robotics)
- 标签: LeRobot
- 创建工具: 该数据集使用 LeRobot 创建
数据集结构
- 代码库版本: v3.0
- 帧率 (FPS): 20
- 总集数 (Episodes): 1
- 总帧数 (Frames): 169
- 总任务数 (Tasks): 1
- 数据文件大小: 100 MB
- 视频文件大小: 200 MB
- 机器人类型: vibeboard_follower_tilt
- 数据拆分: 训练集(train)包含从 0 到 1 的所有数据
特征 (Features)
观测状态 (observation.state)
- 数据类型: float32
- 形状: [7]
- 维度名称: shoulder_pan.pos, shoulder_lift.pos, elbow_flex.pos, wrist_flex.pos, wrist_roll.pos, gripper.pos, tilt.pos
图像观测 (observation.images)
-
base (基础相机):
- 数据类型: video
- 形状: [480, 640, 3] (高, 宽, 通道)
- 视频编码: av1
- 帧率: 20 FPS
- 是否为深度图: 否
-
top (顶部相机):
- 数据类型: video
- 形状: [480, 640, 3]
- 视频编码: av1
- 帧率: 20 FPS
- 是否为深度图: 否
-
wrist (腕部相机):
- 数据类型: video
- 形状: [480, 640, 3]
- 视频编码: av1
- 帧率: 20 FPS
- 是否为深度图: 否
动作 (action)
- 数据类型: float32
- 形状: [7]
- 维度名称: shoulder_pan.pos, shoulder_lift.pos, elbow_flex.pos, wrist_flex.pos, wrist_roll.pos, gripper.pos, tilt.pos
其他特征
- timestamp: float32, 形状 [1]
- frame_index: int64, 形状 [1]
- episode_index: int64, 形状 [1]
- index: int64, 形状 [1]
- task_index: int64, 形状 [1]
数据文件路径
- 数据路径:
data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet - 视频路径:
videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4
引用信息
- 主页: [更多信息待补充]
- 论文: [更多信息待补充]
- BibTeX: [更多信息待补充]



